The Relation Between Earnings Management Using Real ...€¦ · The Relation Between Earnings...

The Relation Between Earnings Management Using

Real Activities Manipulation and Future

Performance: Evidence from Meeting Earnings

Benchmarks*

KATHERINE A. GUNNY, University of Colorado

1. Introduction

Earnings management can be classified into two categories: accruals manage-ment and real activities manipulation (RM). Accruals management involveswithin generally accepted accounting principles (GAAP) accounting choicesthat try to ‘‘obscure’’ or ‘‘mask’’ true economic performance (Dechow andSkinner 2000). RM occurs when managers undertake actions that change thetiming or structuring of an operation, investment, and ⁄or financing transac-tion in an effort to influence the output of the accounting system. Schipper(1989, 92) includes RM in her definition of earnings management anddescribes earnings management as ‘‘a purposeful intervention in the externalfinancial reporting process, with the intention of obtaining some privategain…[a] minor extension of this definition would encompass ‘real’ earningsmanagement, accomplished by timing investment or financing decision toalter reported earnings or some subset of it.’’ This paper examines the extentto which RM is associated with firms just meeting earnings benchmarks.Then, I examine the extent to which RM affects subsequent operatingperformance.

Accruals management is not accomplished by changing the underlyingoperating activities of the firm, but through the choice of accounting meth-ods used to represent those activities. In contrast, RM involves changingthe firm’s underlying operations in an effort to boost current-period earn-ings. Both types of earnings management involve managers’ attempts toincrease ⁄decrease earnings; however, one type affects operations and the

* Accepted by Jeffrey Callen. I am grateful for the guidance and support I have received

from my dissertation chair Xiao-Jun Zhang at the University of California, Berkeley. I

appreciate the helpful comments of Qintao Fan, Sunil Dutta, Maria Nondorf, Shai Levi,

Phil Shane, John Jacob, Naomi Soderstrom, Steve Rock, Joel Demski, Eli Bartov, Baruch

Lev, Paul Zarowin, Thomas Lys, Ronald Dye, Tracey Zhang, Shimon Kogan, Gavin

Cassar, Kin Lo, Bjorn Jorgensen, Brian Burnett, and Qiang Cheng. I am also grateful to

the workshop participants at University of California at Berkeley, New York University,

Northwestern University, University of Florida, Georgia Tech, and University of British

Columbia.

Contemporary Accounting Research Vol. 27 No. 3 (Fall 2010) pp. 855–888 � CAAA

doi:10.1111/j.1911-3846.2010.01029.x

other has no affect on operating activities.1 Examples of RM include over-production to decrease cost of goods sold (COGS) expense and cuttingdesirable research and development (R&D) investments to boost current-period earnings.2

Managers may want to engage in RM versus using accruals manage-ment for several reasons. First, ex post aggressive accounting choices withrespect to accruals are at higher risk for Securities and Exchange Commis-sion (SEC) scrutiny and class action litigation. Second, the firm may havelimited flexibility to manage accruals. For example, accruals management isconstrained by the business operations and accrual manipulation in prioryears (Barton and Simko 2002). Further, accruals management must takeplace at the end of the fiscal year or quarter, and managers face uncertaintyas to which accounting treatments the auditor will allow at that time. Oper-ating decisions are controlled by the manager, whereas accounting choicesare subject to auditor scrutiny. On the other hand, managers may preferaccruals management to RM because accruals management can take placeafter the fiscal year end when the need for earnings management is the mostcertain, whereas RM decisions must be made prior to fiscal year end.

Prior studies provide evidence on the existence of RM (Roychowdhury2006; Baber, Fairfield, and Haggard 1991; Bartov 1993; Bens, Nagar, andWong 2002). The use of RM by managers is supported by Graham, Har-vey, and Rajgopal 2005, who survey 401 financial executives about key fac-tors that drive decisions about reported earnings and voluntary disclosure.They report that 78 percent of the executives interviewed indicated a will-ingness to sacrifice economic value to manage financial reporting percep-tions. Graham et al. (2005, 40) report that ‘‘the opinion of 15 of 20interviewed executives is that companies would ⁄ should take actions such asthese to deliver earnings, as long as the actions are within GAAP and thereal sacrifices are not too large.’’ ‘‘Actions such as these’’ refers to postpon-ing or eliminating expenses (hiring, R&D, advertising, travel, maintenance,

1. Conventional wisdom in prior studies is that managers prefer a higher stock price and

stock price is increasing in earnings (see Fischer and Verrecchia 2000). While the focus

of this study is on income-increasing RM, there are situations in which the manager

may benefit by decreasing earnings. Firms prior to a management buyout, during the

award date of stock options, vulnerable to an antitrust investigation, or seeking import

relief may have incentives to lower reported earnings (e.g., Perry and Williams 1994;

Watts and Zimmerman 1978; Jones 1991).

2. The distinction between cash-based earnings management and RM is that income-

increasing RM will not always affect abnormal cash flow from operations (CFO) and

earnings in the same direction. Reductions of discretionary expenses will lead to abnor-

mally high CFO at the end of the period (assuming discretionary expenses are typically

paid in cash). If a manager engages in overproduction to decrease COGS, the firm will

most likely incur costs on the overproduced items that are not recovered in the current

period through sales which will lead to abnormally low CFO. If the manager engages

in more than one RM method at the same time, then the effect on CFO may be

ambiguous.

856 Contemporary Accounting Research

CAR Vol. 27 No. 3 (Fall 2010)

and capital expenditures to avoid depreciation expense), selling bonds tobook gains, and cutting prices in the fourth quarter. Furthermore, extantempirical accounting literature provides evidence on the existence of RM toachieve various income objectives (see section 2).

Given the existence of RM, I examine the association between RM andfuture performance. In particular, I examine the future operating perfor-mance of firms that use RM to just meet earnings benchmarks. A negativeassociation is consistent with managers using operational discretion to influ-ence the output of the accounting system for managerial rent extraction. Apositive association is consistent with managers using operational discretionto just meet benchmarks in an effort to: (a) attain benefits that allow thefirm to perform better in the future or (b) signal future firm value. Forexample, managers may engage in RM to meet benchmarks in an effort toenhance the firm’s credibility and reputation with stakeholders (Bartov,Givoly, and Hayn 2002; Burgstahler and Dichev 1997). The enhanced repu-tation will enable the firm to perform better in the future because relation-ships with customers, suppliers, and ⁄or creditors are stronger. Alternatively,managers can choose to just meet benchmarks by undertaking RM as away to signal superior future earnings.

The results indicate that, after controlling for size, performance, growthopportunities, and industry, RM (reducing R&D to increase income, reduc-ing selling, general, and administrative (SG&A) expenses to increaseincome, cutting prices to boost sales in the current period, and ⁄or overpro-ducing to decrease COGS expense) is positively associated with firms justmeeting earnings benchmarks. Next, I find firms engaging in RM to justmeet earnings benchmarks have relatively better subsequent performancethan firms that do not engage in RM and miss or just meet the benchmarks.In this particular setting, the results suggest that engaging in RM is notopportunistic, but consistent with the firm attaining current-period benefitsthat allow the firm to perform better in the future or signaling.

Understanding the implications of RM is important not only to stake-holders of the firm, but also to accounting regulators. RM is one potentialconsequence of regulations intended to restrict the discretion in accountingearnings management. For example, through an analytical model, Ewertand Wagenhofer (2005) demonstrate that RM increases when tighteningaccounting standards make accruals management more difficult. Althoughthis study does not specifically address the trade-off between accrualsmanagement and RM, examining the consequences of RM provides generalinformation relevant to assessing the costs and benefits of accounting stan-dards that may interact with the use of RM.

This paper contributes to the literature on earnings management. Byundertaking a comprehensive examination of four types of RM, this paperextends extant research investigating the consequences of earnings manage-ment. Although there are several studies documenting whether RM occurs invarious situations, the existing literature provides little evidence of the effect

Real Activities Manipulation and Future Performance 857


of RM on firms’ subsequent operating performance. This study provides adirect assessment of the impact of RM on future earnings. Examining theimplications of RM on operating performance is important, given the signifi-cance of future performance to the firm and its owners. This paper shows thatusing empirical measures to identify firms that engage in RM to meet zero orlast year’s earnings is incrementally informative about future earnings.

The remainder of the paper is organized as follows. Section 2 discusses thevarious types of RM and presents existing evidence. Section 3 develops testablehypotheses. Section 4 describes the data and methodology. Section 5 presentsthe results and sensitivity analysis. Section 6 provides concluding remarks.

2. Types of RM activities and prior evidence

This study focuses on the following four types of RM demonstrated to existempirically in the prior literature:

(1) decreasing discretionary R&D expense (R&D RM),(2) decreasing discretionary SG&A expense (SG&A RM),(3) timing the sale of fixed assets to report gains (asset RM), and(4) overproduction reflecting an intention to cut prices or extend more

lenient credit terms to boost sales and ⁄or overproduction to decreaseCOGS expense (production RM).

Evidence on RM

Under current accounting rules, R&D expenditures must be charged toexpense as incurred because of the uncertainty of future benefits associatedwith investment in R&D (SFAS No. 2, October 1974).3 As a result, a man-ager interested in boosting current-period income could choose to cutinvestment in R&D, particularly if the realization of the benefit associatedwith the forfeited R&D project impacts the firm in a future period ratherthan the current period. SG&A is included in the analysis because portionsof this expense are similarly subject to managerial discretion. GAAP doesnot recognize intangible assets such as brands, technology, customer loyalty,human capital, and commitment of employees — all of which are createdby expenditures on SG&A — as accounting assets. If the manager decidedto cut employee-training programs intended to increase human capital andcommitment of employees, the economic consequence may not materializein the short term, but in the long term.

Several studies provide evidence that managers cut discretionary spend-ing to achieve earnings targets. Roychowdhury (2006) develops empiricalmeasures to proxy for RM of discretionary expense and reports thatmanagers avoid reporting losses by undertaking RM. Baber et al. (1991)provide evidence that R&D spending is significantly less when spending

3. The Financial Accounting Standards Board (FASB) permits R&D capitalization only

for certain kinds of software (SFAS No. 86).



jeopardizes the ability to report positive or increasing income in the currentperiod. Dechow and Sloan (1991) show that chief executive officers spendrelatively less on R&D in their final years in office. Bushee (1998) providesevidence consistent with institutional investors mitigating this myopicinvestment problem. Bens et al. (2002) show that managers cut R&D andcapital expenditure when faced with earnings per share dilution due to stockoption exercises. Cheng (2004) provides evidence consistent with compensa-tion committees mitigating opportunistic reductions in R&D spending. Theevidence is consistent with managers myopically cutting investment in R&Dto achieve various income objectives.

The timing of asset sales is a manager’s choice, and because gains arereported on the income statement at the time of the sale (the differencebetween the net book value and the current market value), the timing ofasset sales could be used as a way to manage reported earnings. Bartov(1993) provides evidence consistent with managers selling fixed assets toavoid negative earnings growth and debt covenant violations. Herrmann,Inoue, and Thomas (2003) investigate Japanese managers’ use of incomefrom the sale of assets to manage earnings. They find that earnings increase(decrease) through the sale of fixed assets and marketable securities whencurrent operating income falls below (above) management’s forecast ofoperating income.

Sales manipulation refers to the behavior of managers that try toincrease sales during the current year in an effort to increase reported earn-ings. By cutting prices (or extending more lenient credit terms) toward theend of the year in an effort to accelerate sales from the next fiscal year intothe current year, some managers may be willing to sacrifice future profits tobook additional sales this period. The potential costs of sales manipulationinclude loss in future profitability once the firm reestablishes old prices.Managers can manipulate COGS expense in any period by overproducingto spread fixed overhead costs over a larger number of units as long as thereduction in per-unit cost is not offset by inventory holding costs or anyincrease in marginal cost in the current period. Thomas and Zhang (2002)provide evidence consistent with managers overproducing to decreasereported COGS. Roychowdhury (2006) finds evidence that managers usesales manipulation and overproduction in an effort to avoid reportinglosses.

3. Hypothesis development

I examine the relationship between earnings management using RM andfuture performance in situations where managers are more likely to engagein RM. Specifically, I focus on a sample of firms for which earnings man-agement incentives are high. Prior research documents a discontinuityaround zero earnings and last year’s earnings (Hayn 1995; Burgstahler andDichev 1997; Degeorge, Patel, and Zeckhauser 1999; Jacob and Jorgensen2007) and interprets this as evidence of earnings management by firms to



just meet or slightly beat earnings benchmarks. I examine RM in relation tofirms just meeting two earnings benchmarks (zero earnings and last year’searnings). This leads to the following hypothesis:

HYPOTHESIS 1. Firms that just meet ⁄beat earnings benchmarks (zero earn-ings and last year’s earnings) exhibit evidence of real activitiesmanipulation.

Given the existence of RM, I examine whether there are costs associatedwith engaging in various types of RM. Prior literature provides limited evi-dence on whether RM affects future operating performance.4 I examine thesubsequent performance of firms that use RM to just meet earnings bench-marks (zero or last year’s earnings).5 A negative association between justmeeting earnings benchmarks by using RM and subsequent performancesupports prior research that suggests opportunistic managers use accountingor operational discretion to the detriment of shareholders.6 For example,managers could engage in RM to just meet an earnings benchmark toincrease stock prices, job security, or bonuses (Matsunaga and Park 2001).

A positive association between just meeting earnings benchmarks byusing RM and subsequent performance is consistent with two distinctexplanations. First, the act of just meeting the benchmark by engaging inRM may provide benefits to the firm that enables better performance inthe future. For example, Bartov (1993) provides evidence consistent withmanagers selling fixed assets to avoid debt covenant violations. Truemanand Titman (1988) find managers use RM to smooth reported income todecrease the cost of debt. Bartov et al. (2002) suggest that benefits tomeeting earnings expectations may include maximizing stock price, increas-ing management’s credibility for meeting the expectations of stakeholders,

4. Bens et al. (2002) find future performance is relatively lower for firms that cut R&D

expenditures to repurchase shares.

5. When examining the relation between future performance and RM, I assume RM is an

exogenous variable. If RM is endogenously determined such that there is a factor that

affects RM and also affects firms’ future performance (e.g., RM firm-years being repre-

sentative of poor performance), then this study suffers from a potential correlated omit-

ted variable problem. However, I focus on RM conditional on an earnings management

incentive to mitigate the effects of alternative explanations and potential correlated

omitted variables.

6. For example, a manager has the opportunity to undertake a positive net present value

R&D project that requires an initial investment of $100M in period t to generate cash

flows of $80M in both t + 1 and t + 2. In period t, the manager is worried about job

security and ⁄ or the stock price reaction to missing zero earnings, so he rejects the posi-

tive net present value R&D project. In this case, period t earnings are $100M higher;

however, earnings in the subsequent two periods are $80M lower compared to an identi-

cal firm that would have undertaken the R&D project. With respect to production RM,

aggressive price discounts could be used to increase sales volume and allow the manager

to meet zero earnings in the current period; however, cash flows in future periods could

be affected because customers now expect such price discounts.



and avoiding litigation. Graham et al. (2005, 27) find that 86.3 percent ofexecutives ‘‘believe that meeting benchmarks builds credibility with thecapital market’’. Shareholders benefit from managers undertaking RM tojust meet earnings benchmarks to the extent that the benefits exceed thecosts.

Second, the positive association between just meeting earnings bench-marks by engaging in RM and future performance is also consistent withsignaling managerial competence or future firm performance (Bartov et al.2002; Lev 2003).7 Burgstahler and Dichev (1997) suggest that meetingearnings benchmarks may enhance firms’ credibility and reputation withstakeholders such as creditors, suppliers, and customers. Prior literaturereports that firms use discretionary accruals to signal firm value (Subr-amanyam 1996). Graham et al. (2005) find that 74.1 percent of executivestry to meet earnings benchmarks because it helps to convey future growthprospects to investors. Managers may use the joint signal — engaging inRM and just meeting the earnings benchmark — to convey future growthprospects. For example, a manager could choose to meet a benchmark byengaging in RM or miss the benchmark by not engaging in RM. Consis-tent with the signaling explanation, only managers confident in superiorfuture performance will use the joint signal because they expect futureearnings growth to outweigh the adverse impact of using RM and meetingthe benchmark. Firms with relatively worse future performance are notlikely to use the joint signal because investors will be disappointed whenthe firm experiences an impact on earnings from the costs of RM (i.e.,forfeited future cash flows) and the cost of setting earnings expectationshigher by meeting the benchmark in the prior period. Earnings disappoint-ments could lead to impaired management credibility and a higher likeli-hood of litigation.

Alternatively, finding no association between just meeting earningsbenchmarks by engaging in RM and subsequent performance is consistentwith the research design failing to capture RM and ⁄or three other explana-tions. First, no association is consistent with the operational activity labeledas RM being the optimal choice. For example, it could be optimal for themanager to cut a positive net present value R&D project if the benefitsfrom just meeting the earnings benchmark equal the costs of forfeiting theR&D project. In this case, subsequent performance may be insignificantlydifferent from a peer firm. A second alternative explanation could be

7. This explanation does not necessarily imply that shareholders benefit from signaling.

There are potentially less costly alternatives to signaling other than engaging in RM

and just meeting earnings benchmarks. For example, the manager could miss the bench-

mark, but issue a management forecast indicating superior future performance. For

most firms, this may be less costly and therefore a less credible signal of future firm per-

formance. However, for some firms with reputations for providing credible management

forecasts, this could be a costly and effective signal of future performance.



that the consequences of RM are so small that they are undetectable. Forexample, Graham et al. (2005, 40) document chief financial officersadmitting a willingness to engage in RM ‘‘as long as the real sacrifices arenot too large’’. Lastly, it may be that managers engage in RM for severalreasons (e.g., opportunistic, signaling) and the combined effects on futureperformance offset on average. These competing arguments lead to the fol-lowing hypothesis (stated in null form):

HYPOTHESIS 2. There is no association between using RM to just meet ⁄beat earnings benchmarks and future performance.

4. Data and methodology

The sample consists of all firms with available financial data from COMPU-STAT industrial, full-coverage, and research files and stock and size portfo-lio returns from the Center for Research in Security Prices (CRSP). Firmsin the financial industry (SIC 6000–7000) and utility industry (SIC 4400–5000) are excluded because they operate in highly regulated industries withaccounting rules that differ from other industries. The sample includesannual data for firms covering the years from 1988 to 2002. The sample isrestricted to pre-2003 data, so there are several years of subsequent earn-ings to examine. The sample is restricted to post-1987 data because dataon income from asset sales are not available on COMPUSTAT before1987.

The R&D RM sample contains all firm-years with nonzero R&Dexpense data and the COMPUSTAT variables necessary to estimate abnor-mal R&D expense (28,308 observations and 4,028 firms). The SG&A RMsample contains all firm-years with nonzero SG&A expense data and theCOMPUSTAT variables necessary to estimate abnormal SG&A expense(46,156 observations and 6,021 firms). The asset RM sample consists of allfirm-years with the COMPUSTAT variables necessary to estimate abnormalgain on asset sales (33,528 observations and 5,452 firms). The productionRM sample consists of all firm-years with nonzero inventory and COGSdata, and the COMPUSTAT variables necessary to estimate abnormal pro-duction costs (39,432 observations and 5,526 firms).

Identification of RM

Given the inherent difficulty in identifying earnings management withoutknowing the manager’s true intention, one criticism of the literature is thatany earnings management identified may be a result of an omitted variableor may be capturing behavior other than intentional manipulation. Thiscriticism applies to my study; however, I try to mitigate these concerns inseveral ways. First, I draw on prior literature to develop models to estimatethe expected (i.e., ‘‘normal’’) level of the operational activities associatedwith RM. Second, to distinguish between the two scenarios described above,



I examine a setting where the manager is more likely to engage in RM.Specifically, I focus on firms just meeting zero and last year’s net income.8

Measurement of RM

The normal level of R&D expense is estimated using the following model:

RDt

At�1¼ a0 þ a1

1

At�1þ b1MVt þ b2Qt þ b3

INTt

At�1þ b4

RDt�1

At�1þ eR&D

t ð1Þ;

where (COMPUSTAT data items in brackets):

RD = R&D expense [Data46],A = total assets [Data6],MV = the natural log of market value [Data199*Data25],Q = Tobin’s Q [((Data199*Data25) + Data130 + Data9 + Data34) ⁄

Data6], andINT = internal funds [Data18 + Data46 + Data14].

Equation (1) is based on prior research (Berger 1993; Roychowdhury2006) that develops an expectations model for the level of R&D intensity.The model is estimated for every year (1988–2000) and industry (two-digitSIC). The independent variables are designed to control for factors thatinfluence the level of R&D spending. I use the natural logarithm of themarket value of equity (MV) to control for size. Tobin’s Q is a proxy forthe marginal benefit to marginal cost of installing an additional unit of anew investment. Internal funds (INT) are a proxy for reduced funds avail-able for investment. The prior year’s R&D (RDt )1) serves as a proxy forthe firm’s R&D opportunity set and the coefficient would be expected to bepositive.

The normal level of SG&A is estimated using the following model:

SGAt

At�1¼ a0þa1

1

At�1þb1MVtþb2Qtþb3

INTt

At�1þb4

DSt

At�1þb5

DSt

At�1�DDþ eSG&A

t

ð2Þ;


SGA = SG&A [Data189],A = total assets [Data6],MV = the natural logarithm of market value [Data199*Data25],

8. I do not focus on analysts’ forecasts for two reasons: (1) RM must take place before

the end of the year and managers are unlikely to know what the analysts’ forecast of

earnings will be prior to the earnings announcement and (2) Matsumoto (2002) exam-

ines the mechanisms managers use to avoid missing analysts’ forecasts and finds evi-

dence consistent with forecast guidance dominating accruals manipulation as a

mechanism for avoiding negative surprises. Therefore, it is unclear whether using firms

that just meet the analysts’ forecast would increase the power of correctly identifying

RM.



Q = Tobin’s Q [((Data199*Data25) + Data130 + Data9 + Data34) ⁄Data6],

INT = internal funds [Data18 + Data46 + Data14],S = total sales [Data12], andDD = indicator variable equal to 1 when total sales decrease between t ) 1

and t, zero otherwise.

Equation (2) is similarly estimated by year and industry. In addition tomarket value, Tobin’s Q, and internal funds, I incorporate controls for‘‘sticky’’ cost behavior (Anderson, Banker, and Janakiraman 2003). Costs aresticky if the magnitude of a cost increase associated with increased sales isgreater than the magnitude of a cost decrease associated with an equaldecrease in sales. The general theory is that managers trade off the expectedcosts of maintaining unutilized resources during periods of weak demandwith the expected adjustment costs of replacing these resources if demand isrestored. As a result, I use change in sales times an indicator variable equal toone when sales revenue decreases between t ) 1 and t (DSt *DDt). Not includ-ing this element in the SG&A expectations model may lead to underestimat-ing (overestimating) the response of costs to increases (decreases) in sales.9

The normal level of gain on asset sales is estimated using the followingmodel:

GainAt

At�1¼ a0þa1

1


INTt

At�1þb4

ASalest

At�1þb5

ISalest

At�1þ eAsset

t

ð3Þ;


GainA = income from asset sales [Data213*()1); note: Data213 is codednegative for gains and positive for losses by COMPUSTAT],

A = total assets [Data6],MV = the natural logarithm of market value [Data199*Data25],Q = Tobin’s Q [((Data199*Data25) + Data130 + Data9 + Data34) ⁄

Data6],INT = internal funds [Data18 + Data46 + Data14],ASales = long-lived assets sales [Data107], andISales = long-lived investment sales [Data109].

Equation (3), estimated by year and industry, is based on Bartov 1993and augmented by variables in Herrmann et al. 2003 shown to influence thelevel of gain on asset sales. Market value is included to control for sizeeffects. Internal funds control for reduced funds available for investmentand Tobin’s Q is a proxy for the marginal benefit to marginal costof installing an additional unit of a new investment, both of which may

9. This sticky cost behavior has only been shown with respect to SG&A; therefore, I only

include change in sales and change in sales times a decrease dummy in model 2.



influence the decision to sell fixed assets. Introducing asset sales as anexplanatory variable in (3) requires that the relation between income fromasset sales (GainA) and asset sales (ASales) and investment sales (ISales) bemonotonic. Therefore, the variables are transformed to make the relation-ship monotonic, so when income from asset sales is negative, asset sales andinvestment sales enter the regression with negative signs. Thus, a positivecoefficient is expected. Consistent with prior literature (Bartov 1993; Herr-mann et al. 2003), I interpret high residuals from model 3 as indicative ofasset sales manipulation.10

The normal level of production cost is estimated using the followingmodel:

PRODt

At�1¼ a0þa1

1


St

At�1þb4

DSt

At�1þb5

DSt�1

At�1þ eProduction

t

ð4Þ;


PROD = COGS plus change in inventory [Data41 + Data303],A = total assets [Data6],MV = the natural log of market value [Data199*Data25],Q = Tobin’s Q [((Data199*Data25) + Data130 + Data9 + Data34) ⁄

Data6], andS = sales [Data12].

Model 4 is estimated by year and industry. The model is based onDechow et al. 1998 and Roychowdhury 2006 to estimate the normal levelof production. I augment their regression by including market value andTobin’s Q.11 Sales, change in sales, and lagged change in sales are includedto control for any product demand changes that might directly influence thelevel of production. Abnormally high production costs for a given saleslevel are indicative of either sales manipulation due to abnormal price dis-counts or COGS expense manipulation by overproduction (Roychowdhury

10. I employ alternative expectations models for R&D expense, SG&A expense, and gain

(loss) on asset sales. First, R&D and SG&A expense (divided by assets) are modeled

solely as a function of sales, as described by Dechow, Kothari, and Watts 1998. Second,

the normal level of income from asset sales is estimated as income from asset sales

minus the median for the corresponding industry and year. The results for these rela-

tively simpler models are qualitatively similar.

11. Production costs have not shown the same sensitivity to internal funds as discretionary

expense and asset sales. For example, if the firm is cash constrained, decreasing discre-

tionary investment will increase cash flow from operations and selling fixed assets will

increase cash flow from investing. Engaging in production RM will lead to relatively

lower cash flow in the current period, but higher cash flow in the next period because

sales in t + 1 were moved to t (in the case of Sales RM) and firms can use excess pro-

duction from t in t + 1 (in the case of COGS RM). Therefore, I do not include INT in

the model.



2006). Therefore, I use abnormal production costs as one proxy for salesmanipulation and ⁄or COGS manipulation.12

Firms suspected of RM have abnormal levels (i.e., residuals) from mod-els 1–4 in the quintile consistent with RM. Firms suspected of R&D(SG&A) RM are firms in the lowest quintile of abnormal R&D (SG&A)expense. Firms suspected of Asset (Production) RM are firms in the highestquintile of abnormal gain on asset sales (production costs).

Incentive to engage in earnings management

To identify firms that just meet zero earnings, I group firm-years intointervals based on net income (Data172) divided by total assets (Data6) atthe beginning of the year.13 Then, I construct categories of scaled earningsfor widths of 0.01. I identify firms that just meet zero earnings by concen-trating on firm-years in the interval to the immediate right of zero. Thefirms to the immediate right of zero have net income scaled by total assetsthat is greater than or equal to zero, but less than 0.01 (MEET_ZERO).Similarly, to identify firms that just meet zero earnings growth, I groupfirm-years into intervals based on the change in net income divided bytotal assets at the beginning of the year. Then, I construct categories ofscaled changes in earnings for widths of 0.01. The firms to the immediateright of zero have earnings scaled by total assets that are greater than orequal to zero, but less than 0.01 (MEET_LAST). I identify firms that aresuspected of engaging in earnings management to just meet zero earningsor last year’s earnings as firm-years that fall within either interval(BENCH).14,15

I construct additional classifications based on the scaled earnings (andchange in earnings) intervals to facilitate the comparison of BENCH firmsto non-BENCH firms. From the sample of firms not classified as BENCH,I classify firms where scaled net income (or change in net income) isgreater than or equal to 0.01 as BEAT firms, greater than or equal to)0.01 but less than zero (and not classified as BEAT) as JUSTMISS

12. To mitigate the confounding influence of accruals management, I analyze production

costs instead of COGS expense (or change in accounts receivable). For example, if a

manager decided to postpone the write-down of obsolete inventory in an effort to

decrease reported COGS, this action would manifest as abnormally low COGS expense.

Using COGS as the RM proxy would misclassify accruals management as RM. By

examining production costs (COGS + DINV), the manager’s action would not affect

production costs because the change in inventories would be correspondingly higher to

offset lower COGS. Similarly, it would be difficult to parse out the effects of RM versus

accruals management when using change in accounts receivable as an RM proxy.

13. The results are qualitatively similar using net income before special items and pre-tax

income.

14. The inferences do not change using MEET_ZERO and MEET_LAST separately.

15. Using the Z-statistic described in footnote 6 of Burgstahler and Dichev 1997, the fre-

quency of firms in the bins just to the right of zero (MEET_ZERO and MEET_LAST)

are statistically different from the expected frequency.



firms, and less than )0.01 (and not classified as BEAT or JUSTMISS) asMISS firms.

Descriptive statistics

Table 1 reports the estimation results for (1) through (4). For every industry-year with more than 15 firms, the equations are estimated cross-sectionallyover the period from 1988 to 2002. All variables are winsorized at the top andbottom 1 percent of their distributions to avoid the influence of outliers. Thereported coefficients are the mean value of the coefficients across industry-years. p-values are calculated using the standard error of the mean coefficientsacross industry-years (Fama and Macbeth 1973). The reported observationsand adjusted R2 are means across industry-years. The coefficient estimates aresignificant and with predicted signs. One exception is that SG&A in (2) doesnot exhibit sticky cost behavior as predicted by Anderson et al. 2003. TheR&D expense equation has the highest average adjusted R2, 0.86 acrossindustry-years. The gain on asset equation has the lowest average adjustedR2, 0.28 across industry-years. The equations seem to have reasonable explan-atory power and the adjusted R2s are consistent with prior literature.

Table 2, panel A shows descriptive statistics related to the residualsfrom (1) through (4). To limit the influence of outliers, all continuous vari-ables are winsorized at the top and bottom 1 percent of their distributionfor presentation in Table 2 and implementation of model 5. The mean(median) residual from the R&D model is 0.00 ().001). The mean totalassets and R&D expense for the sample are 1,338 and 65 million, respec-tively (untabulated). Therefore, on average, the median level of abnormalR&D is 1.2 million below normal levels for firms in comparable industries,which is about 1.5 percent of average total assets. The distributions of theresiduals tend to exhibit properties consistent with the normal distribution.The skewness data for all the distributions are relatively close to zero, sug-gesting that the distributions are symmetrically distributed. The kurtosisdata for model 3 suggests that the tails of the distribution are heavier thanfor a normal distribution, which is consistent with firms engaging in assetRM (and moving into the tails).

Table 2, panel B reports Pearson correlations between the RM residualsand other firm characteristics. The correlation matrix reveals that the R&Dresiduals are negatively correlated with SIZE, ROA, and CFO. The SG&Aresiduals are significantly related to SIZE (negative) and ROA (positive).The asset residuals are not significantly related to any of the control vari-ables. The production residuals are significantly related to SIZE (positive)and ROA (negative). The R&D residuals are positively correlated with theSG&A and Production residuals. This suggests that, while managers mayengage in R&D and SG&A RM simultaneously, they do not engage inR&D and Production RM simultaneously. The overlap in the number offirms suspected of engaging in R&D and SG&A RM is 28.4 percent, andR&D and production RM is 20.4 percent (untabulated). The correlation



TABLE

1

Estim

ationofthenorm

allevel

ofR&D

expense,SG&A

expense,gain

onassetssales,andproductioncosts

Model

1:R&D

RD

t⁄A

t)

1

Model

2:SG&A

SGA

t⁄A

t)

1

Model

3:Asset

Gain

GainA

t⁄A

t)

1

Model

4:Production

PROD

t⁄A

t)

1

Intercept

)0.006(0.00)***

Intercept

0.297(0.00)***

Intercept

0.000(0.60)

Intercept

)0.174(0.00)***

1A

t�1

0.071(0.00)***

1A

t�1

2.364(0.00)***

1A

t�1

0.013(0.34)

1A

t�1

0.907(0.17)

MVt

0.001(0.01)***

MVt

)0.015(0.00)***

MVt

0.000(0.24)

MVt

0.019(0.00)***

Qt

0.002(0.00)***

Qt

0.033(0.00)***

Qt

)0.001(0.00)***

Qt

)0.070(0.00)***

INT

t

At�

10.019(0.00)***

INT

t

At�

10.125(0.00)***

INT

t

At�

10.005(0.00)***

St

At�

10.799(0.00)***

RD

t�1

At�

10.897(0.00)***

DS

t

At�

10.165(0.00)***

ASal

est

At�

10.250(0.00)***

DS

t

At�

10.132(0.00)***

DS

t

At�

1�

DD

0.000(0.16)

ISal

est

At�

10.012(0.83)

DS

t�1

At�

1)0.046(0.00)***

No.of

industry

)year

342

550

457

510

Avg.no.

ofobs.

83

84

74

77

Adj.R2

0.86

0.40

0.28

0.82

Notes:

Thefollowingordinary

least

squaresregressionsare

estimatedcross-sectionallywithin

each

industry

(two-digitSIC

)andyearfrom

1988

to2002withatleast

15observations.Thereported

coefficients

are

themeanvalueofthecoefficients

across

theindustry-years.

Two-tailed

p-values

(inparentheses)are

calculatedusingthestandard

errorofthemeancoefficients

across

theindustry-years.

Theadjusted

R2andthenumber

ofobservationsisthemeanacross

theindustry-years.Thevariablesare

defined

asfollows

(COMPUSTAT

data

item

sin

brackets):

RD

=R&D

expense

[Data46]

(Thetable

iscontinued

onthenextpage.)



TABLE

1(C

ontinued)

A=

totalassets[D

ata6]

MV

=thenaturallogofmarket

value[D

ata199*Data25]

Q=

Tobin’sQ

[((D

ata199*Data25)+

Data130+

Data9+

Data34)⁄D

ata6]

INT

=internalfundsdivided

bylagged

totalassets[D

ata18+

Data46+

Data14]

SGA

=SG&A

expense

[Data189]

S=

totalsales[D

ata12]

DD

=indicatorvariable

equalto

1when

sales[D

ata12]decreasesbetweent)1andt,zero

otherwise

GainA

=incomefrom

asset

sales[D

ata213*(-1);note:Data213iscoded

negativeforgainsandpositiveforlosses]

ASales=

long-lived

asset

sales[D

ata107]

ISales

=long-lived

investm

entsales[D

ata109]

PROD

=COGSpluschangein

inventory

[Data41+

DData303]



between the SG&A residual and the production residual is very high()0.5405). Interestingly, 52.1 percent of firms in the lowest SG&A residualquintile are also in the highest production residual quintile (untabulated).

TABLE 2

Descriptive statistics

Panel A: Descriptive statistics of residuals from models 1–4

Mean Median Std.dev.1st

quartile3rd

quartile Skewness Kurtosis

R&D

residuals

0.000 )0.001 0.07 )0.02 0.01 1.53 8.66

SG&A

residuals

0.000 )0.018 0.26 )0.13 0.10 0.75 2.82

Gain Asset

residuals

0.000 )0.001 0.01 0.00 0.00 3.61 20.95

Production

residuals

0.000 )0.006 0.25 )0.14 0.11 0.34 1.93

Panel B: Pearson correlation matrix

SIZE MTB ROA CFO

R&D

Residual

SG&A

Residual

Asset

Residual

MTB )0.027***ROA 0.012*** )0.002CFO 0.003 0.000 0.042***

R&D

residual

)0.013** )0.008 )0.042***)0.039***

SG&A

residual

)0.021*** )0.001 0.009** 0.003 0.1135***

Asset

residual

0.004 0.000 0.000 0.000 )0.0357***)0.0059

Production

residual

0.052*** 0.006 )0.008* )0.003 0.0241***)0.5405*** 0.0171***

CFO = cash flow from operations divided by lagged total assets

Notes:

* ⁄ ** ⁄ *** represent statistical significance at 10 percent ⁄ 5 percent ⁄ 1 percent levels,

two-tailed. Firm-years from 1988 to 2002. RM residuals are estimated from

models 1–4. See Table 1 for estimation and variable definitions. The variables

are defined as follows:

SIZE = the natural logarithm of total assets

MTB = the market value of equity divided by the book value of equity

ROA = income before extraordinary items divided lagged total assets



Thus, it appears many firms simultaneously engage in both SG&A and pro-duction RM, which may explain the high correlation.

5. Results

Abnormal RM and just meeting zero earnings and last year’s earnings

To test the association between firms just meeting benchmarks and RM(Hypothesis 1), I estimate the following equation:

Abnormal RMt ¼ c0 þ c1BENCHt þ c2SIZEt þ c3MTBt þ c4ROAt þ et ð5Þ;

where:

BENCH = an indicator variable that is set equal to one if (a) net incomedivided by total assets is between 0 and 0.01 or (b) the changein net income divided by total assets between t ) 1 and t isbetween 0 and 0.01, zero otherwise,

SIZE = the natural logarithm of total assets,MTB = the market value of equity divided by the book value of equity,

andROA = income before extraordinary items divided by lagged total

assets.

Equation (5) is estimated using four measures of Abnormal RM as thedependent variable: abnormal R&D expense (Abnormal R&D), abnormalSG&A expense (Abnormal SG&A), abnormal gain on asset sales (AbnormalGainAsset), and abnormal production costs (Abnormal Production).16 BothAbnormal GainAsset and Abnormal Production are multiplied by ()1) so thatlower values are consistent with RM. SIZE controls for size effects andMTB controls for growth opportunities. ROA is included to address con-cerns that RM is correlated with performance. Because the error terms arelikely to exhibit cross-sectional correlation and auto correlation, I estimatepooled regressions and compute the t-tests using Roger’s robust standarderrors, correcting for firm clusters (Petersen 2009).

Table 3 reports the results from the estimation of (5). Abnormal R&D isnegatively associated with firms that just meet zero or last year’s earnings(coefficient )0.0035, p-value < 0.05). The coefficient on BENCH whenAbnormal SG&A is the dependent variable is )0.0099 and significant at a 5percent level. The results for discretionary expense suggest firms engage inRM of R&D and SG&A expense to just meet zero and last year’s earnings.The coefficient on BENCH when Abnormal GainAsset is the dependentvariable is not significantly different from zero. It appears firms that just

16. One criticism of this model could be that the independent variables (SIZE, MTB, and

ROA) control for the same variations controlled for in models 1 through 4; therefore, a

univariate analysis may be appropriate. I keep the control variables used in Roy-

chowdhury 2006 to facilitate comparison between the studies. The univariate results are

qualitatively similar.



TABLE 3

Cross)sectional regressions relating abnormal residuals to firms just meeting zero or

last years earnings

Abnormal RMt ¼ c0 þ c1BENCHt þ c2SIZEt þ c3MTBt þ c4ROAt þ etþ1 (5)

VariableAbnormalR&Dt

AbnormalSG&At

AbnormalGainAssett

*()1)

AbnormalProductiont

*()1)

AbnormalAggregateRMt(R&D+ SG&A

+ Production)

Intercept 0.0041 0.0123 0.000 0.038 0.109

(2.45)** (1.48) (1.39) (2.44)*** (4.34)***

BENCHt )0.0035 )0.0099 0.00001 )0.048 )0.044()2.14)** ()1.98)** (0.04) ()1.97)** ()2.64)***

SIZEt )0.0007 )0.0034 )0.00003 )0.006 )0.014()2.32)** ()2.54)** ()0.83) ()1.87)* ()2.95)***

MTBt )0.0002 0.0011 0.000 )0.001 0.001

()1.46) (1.93)** ()2.00)** ()1.64) (0.39)

ROAt )0.0004 0.0000 0.000 0.000 0.103

()3.06)*** (0.45) ()0.08) (1.28) (4.53)***

# Obs. 27,613 44,960 32,715 38,394 24,402

# Firms 4,003 5,985 5,412 5,489 3,744

Adj. R2 0.0029 0.0013 0.0003 0.0029 0.0141

ROA = income before extraordinary items divided lagged total assets

Notes:

* ⁄ ** ⁄ *** represent statistical significance at 10 percent ⁄ 5 percent ⁄ 1 percent

levels, two-tailed. Sample consists of firm-years from 1988 to 2002. The

t-tests are computed using Roger’s robust standard errors correcting for

firm clusters. The coefficient estimates are from ordinary least squares

regressions relating the residuals from models 1–4 to an indicator variable for

whether the firm just meets zero earnings or last year’s earnings and control

variables. Both Abnormal GainAsset and Abnormal Production are multiplied

by (-1) so that lower values are consistent with RM. The variables are

defined as follows:

BENCH = an indicator variable equal to one if(a) net income divided by total

assets is greater than or equal to 0 but less than 0.01, or(b) the change

in net income divided by total assets between t ) 1 and t is greater than

or equal to 0 but less than 0.01, zero otherwise

SIZE = the natural logarithm of total assets

MTB = the market value of equity divided by the book value of equity



meet the earnings benchmarks are not associated with abnormally high gainon asset sales.17 The coefficient on BENCH when Abnormal Production isthe dependent variable is )0.048 and is significant at a 5 percent level. There-fore, firms just meeting earnings benchmarks exhibit higher production costs,which is consistent with these firms engaging in production RM.18

Because firms might engage in more than one type of RM simulta-neously, I aggregate the three RM measures shown to be associated withjust meeting zero and last year’s earnings (Abnormal R&D, AbnormalSG&A, and Abnormal Production). Abnormal Aggregate RM is the sum ofthe residuals from the R&D model 1, SG&A model 2, and productionmodel 3 multiplied by )1. The last column in Table 3 shows the resultsfrom the estimation of (5) using the Abnormal Aggregate RM measure asthe dependent variable. The coefficient on BENCH is )0.044 and is signifi-cant at a 1 percent level. Consistent with prior literature, the resultsreported in Table 3 indicate that managers engage in R&D, SG&A, andproduction RM to just meet earnings benchmarks.

Abnormal RM and future performance

While it appears that managers engage in RM to just meet the earningsbenchmarks, ex ante it is unclear whether this behavior will have an eco-nomically significant association with future performance. In this section, Iexamine the extent to which RM affects subsequent performance. Table 4provides descriptive statistics of industry-adjusted ROA (AdjROA) preced-ing, including and subsequent to year t by earnings and RM categories.19

AdjROA equals the difference between firm-specific ROA and the medianROA for the same year and industry (two-digit SIC). AdjROA and assetsare winsorized at the top and bottom 1 percent of their distributions forpresentation in Table 4. For the R&D, SG&A, and Production samples,about 4 percent of all firm-years just meet an earnings benchmark (1,118,

17. One issue with identifying asset sales manipulation in this way is that it is difficult to

argue that firms making abnormally high profit from selling assets are engaging in RM.

Therefore, as a robustness check, like Zang 2007, I estimate asset RM firm-years as (a)

firms with positive income from asset sales (GainA) and (b) firms with small residuals

from model 3. I find qualitatively similar results when defining asset RM this way — an

insignificant association between asset RM residuals and just meeting the earnings

benchmark (coefficient 0.001, t = 0.92).

18. Because production RM reflects two types of RM and COGS RM should only be avail-

able to firms in the manufacturing industry, I estimate model 5 excluding all nonmanu-

facturing firms. For this subsample, the coefficient on BENCH is 0.003 (untabulated)

and significantly negative (p < 0.01). Therefore, the results are robust to the manufac-

turing sample.

19. I use industry-adjusted performance measures to control for differences in industry con-

centration that may affect the performance measure. I examine the robustness of the

results to using net income plus interest expense (to isolate the effects of financing) as

the performance measure. The association between RM and future performance are

qualitatively similar using this measure.



TABLE

4

Descriptivestatisticsoffirm

sbyearnings(changein

earnings)

category

andRM

#

Mean

Assets

(millions)

AdjR

OA

t)

2AdjR

OA

t)

1AdjR

OA

tAdjR

OA

t+

1AdjR

OA

t+

2

R&D

sample

Allfirm

s27,613

1,833

)6.2

)6.8

)6.5

)3.3

)2.7

BEAT

17,014

2,240

3.9

5.5

9.1

5.2

4.0

BENCH

1,118

3,606

)0.6

)1.9

)0.6

)2.7

)1.3

JUSTMISS

553

2,786

)2.4

)2.2

)1.5

)2.6

)2.5

MISS

8,928

776

)28.0

)31.7

)37.2

)20.5

)16.6

R&D

RM

5,517

3,479

)12.9

)13.6

)5.6

)4.0

)3.6

BENCH

*R&D

RM

185

6,852

)5.2

)8.7

0.2

)1.4

)1.2

BENCH

(noR&D

RM)

933

1,938

0.3

)0.6

)0.7

)2.9

)1.3

SG&A

sample

Allfirm

s44,960

1,655

)3.3

)3.9

)3.9

)1.7

)1.3

BEAT

29,431

1,914

3.8

5.2

8.2

4.8

3.7

BENCH

2,049

3,093

)0.3

)1.7

)1.0

)2.0

)1.4

JUSTMISS

1,008

2,354

)1.5

)1.7

)2.2

)2.7

)2.5

MISS

12,472

751

)21.9

)26.2

)33.1

)17.5

)13.8

SG&A

RM

8,987

1,105

)12.6

)12.9

)10.3

)6.2

)5.7

BENCH

*SG&A

RM

353

1,211

)2.8

)4.8

)0.2

)1.0

)1.9

BENCH

(noSG&A

RM)

1,696

3,484

0.2

)1.1

)1.1

)2.3

)1.2

Productionsample

Allfirm

s38,394

1,754

)3.5

)3.9

)2.8

)1.4

)1.0

BEAT

25,099

2,016

3.8

5.2

7.8

4.8

3.8

BENCH

1,756

3,393

)0.4

)1.7

)0.9

)2.0

)1.4

(Thetable

iscontinued

onthenextpage.)



TABLE

4(C

ontinued)

#

Mean

Assets

(millions)

AdjR

OA

t)

2AdjR

OA

t)

1AdjR

OA

tAdjR

OA

t+

1AdjR

OA

t+

2

JUSTMISS

857

2,577

)1.6

)1.8

)1.8

)2.4

)2.0

MISS

10,682

802

)22.5

)26.2

)28.2

)16.7

)13.1

ProductionRM

7,671

3,260

)11.4

)13.3

)12.3

)7.3

)5.6

BENCH

*ProductionRM

371

9,182

)2.8

)3.5

)1.0

)1.4

)1.2

BENCH

(noProductionRM)

1,385

1,842

0.2

)1.3

)0.9

)2.2

)1.5

ProductionRM

=anindicatorvariable

equalto

oneiftheresidualfrom

productionmodel

4isin

thehighestquintile,zero

otherwise

Notes:

Sample

consistsoffirm

-years

from

1988to

2002.RM

residualsare

estimatedfrom

(1)–(4).Thevariablesare

defined

asfollows:

ROA

=incomebefore

extraordinary

item

sdivided

lagged

totalassets

AdjROA

=thedifference

betweenfirm

-specificROA

andthemedianROA

forthesameyearandindustry

(two-digitSIC

)

BENCH


equalto

oneif(a)net

incomedivided

bytotalassetsisgreaterthanorequalto

0butless

than0.01,or(b)the

changein

net

incomedivided

bytotalassetsbetweent

)1andtisgreaterthanorequalto

0butless

than0.01,zero

otherwise

BEAT


equalto

oneif(a)net

incomedivided


0.01,or(b)thechangein

net

incomedivided

bytotalassetsbetweent)1andtisgreaterthanorequalto

0.01and(c)BENCH

notequalto

one,

zero

otherwise

JUSTMISS


equalto

oneif(a)net

incomedivided


)0.01butless

than0,or(b)the

changein

net

incomedivided



)0.01butless

than0and(c)BENCH

or

BEAT

isnotequalto

one,

zero

otherwise

MISS


thatissetequalto

oneif(a)net

incomedivided

bytotalassetsisless

than

)0.01,or(b)thechangein

net

incomedivided


)1andtisless

than

)0.01and(c)BENCH,BEAT

orJUSTMISSisnotequalto

one,

zero

otherwise

R&D

RM


equalto


theR&D

model

1isin

thelowestquintile,zero

otherwise

SG&A

RM


equalto


theSG&A

model

2isin


otherwise



2,049, and 1,756, respectively). On average, firm-years around the bench-mark range (BENCH, JUSTMISS) perform better than MISS firms butworse than BEAT firms. The last three rows of each panel show the perfor-mance of RM firms, RM firms that just meet an earnings benchmark, andnon-RM firms (i.e., those firms in the four quintiles not consistent withRM) that just meet an earnings benchmark. For all three samples, firm-years in the residual quintile consistent with RM perform better in t + 1and t + 2 than in the previous three years. For the R&D and Productionsamples, it appears that firms that just meet an earnings benchmark byusing RM have higher subsequent AdjROA than firms that just meet anearnings benchmark but do not engage in RM. For the SG&A sample, itappears that BENCH firms that engage in RM have higher AdjROA thannon-RM BENCH firms in year t + 1, but not t + 2.

Interpreting the results of the univariate analysis is difficult due to sys-tematic variation in future ROA with current performance, size, market-to-book, returns and the probability of bankruptcy. To test whether thereis an association between using RM to just meet earnings benchmarksand future performance (Hypothesis 2), I estimate the following equation:

AdjROAtþi or AdjCFOtþi ¼ c0 þ c1BEATt þ c2JUSTMISSt þ c3BENCHt

þ c4RMt þ c5BENCH�RMt þ c6AdjROAt

þ c7SIZEt þ c8MTBt þ c9RETURNt

þ c10ZSCOREt�1 þ etþ1 ð6Þ;

where:

i = 1, 2, 3,ROA = income before extraordinary items divided by lagged total

assets,AdjROA = industry-adjusted ROA equals the difference between firm-

specific ROA and the median ROA for the same year andindustry (two-digit SIC),

CFO = CFO divided by lagged total assets,AdjCFO = industry-adjusted CFO equals the difference between firm-

specific CFO and the median CFO for the same year andindustry (two-digit SIC),

BENCH = an indicator variable that is set equal to one if (a) netincome divided by total assets is between 0 and 0.01, or (b)the change in net income divided by total assets between t )1 and t is between 0 and 0.01, zero otherwise,

BEAT = an indicator variable equal to one if (a) net income dividedby total assets is greater than or equal to 0.01 or (b) thechange in net income divided by total assets between t ) 1and t is greater than or equal to 0.01 and (c) BENCH notequal to one, zero otherwise, and



JUSTMISS = an indicator variable equal to one if (a) net incomedivided by total assets is greater than or equal to )0.01but less than 0 or (b) the change in net income dividedby total assets between t ) 1 and t is greater than orequal to )0.01 but less than 0 and (c) BENCH orBEAT is not equal to one, zero otherwise;

where RM:

R&D RM = an indicator variable equal to one if the residual from theR&D model 1 is in the lowest quintile, zero otherwise,

SG&A RM = an indicator variable equal to one if the residual from theSG&A model 2 is in the lowest quintile, zero otherwise,

Production RM = an indicator variable equal to one if the residual fromthe production model 4 is in the highest quintile, zerootherwise,

Aggregate RM = an indicator variable equal to one if the sum of the residu-als from the R&D model 1, SG&A model 2 and produc-tion model 3*)1 is in the lowest quintile, zero otherwise,

SIZE = the natural logarithm of total assets,MTB = the market value of equity divided by the book value of

equity,RETURN = size adjusted abnormal returns computed as the

monthly buy and hold raw return minus the monthlybuy and hold return on a size matched decile portfolio offirms compounded over 12 months of fiscal year t, and

ZSCORE = a measure of financial health computed as: 3.3*(Netincomet ⁄Assetst ) 1) + 1.0*(Salest ⁄Assetst-1) + 1.4*(Retained Earningst ⁄Assetst-1) + 1.2*(Working Capitalt ⁄Assetst )1)

SIZE controls for size effects and MTB controls for growth opportuni-ties. In the context of R&D and SG&A, controlling for the life cycle (i.e.,MTB) is important given the ‘‘maturity hypothesis’’, which predicts that asfirms mature they experience a decline in their investment opportunity set. Iinclude AdjROA to control for the time series properties of performance. Ialso include RETURN to control for the association between stock perfor-mance and future earnings (Kothari and Sloan 1992). ZSCORE is a modi-fied version of Altman’s Z-score (Mackie-Mason 1990) and is used tocontrol for the financial health of the firm. All continuous variables arewinsorized at the top and bottom 1 percent of their distribution to limit theinfluence of outliers for presentation in Table 5 and implementation ofmodel 6. The intercept (c0) represents the average performance of firms thatdo not use RM and miss the earnings benchmark by more than 0.01.

Hypothesis 2 focuses on firms engaging in RM to just meet earningsbenchmarks beyond the broadened focus on all firms engaging in RM.



TABLE5

Pearsoncorrelationmatrix

R&D

sample

(N=

23,041)

SG&A

sample

(N=

36,501)

Productionsample

(N=

31,855)

BEAT

MISS

JUST

MISS

AdjROA

SIZ

EMTB

RETURN

BENCH

R&D

RM

BENCH*

RM

BENCH

SG&A

RM

BENCH*

RM

BENCH

Production

RM

BENCH*

RM

RM

)0.018

(0.003)

)0.015

(0.001)

0.006

(0.217)

BENCH

*RM

0.400

(<.0001)

0.164

(<.0001)

0.407

(<.0001)

0.178

(<.0001)

0.451

(<.0001)

0.198

(<.0001)

BEAT

)0.260

(<.0001)

)0.023

(<.0001)

)0.104

(<.0001)

)0.301

(<.0001)

)0.069

(<.0001)

)0.122

(<.0001)

)0.284

(<.0001)

)0.027

(<.0001)

)0.131

(<.0001)

MISS

)0.142

(<.0001)

0.039

(<.0001)

)0.057

(<.0001)

)0.135

(<.0001)

0.085

(<.0001)

)0.055

(<.0001)

)0.140

(<.0001)

0.022

(<.0001)

)0.064

(<.0001)

)0.875

(<.0001)

JUSTMISS

)0.030

(<.0001)

)0.024

(<.0001)

)0.012

(0.048)

)0.033

(<.0001)

)0.015

(0.001)

)0.014

(0.004)

)0.032

(<.0001)

0.007

(<.0001)

)0.015

(0.008)

)0.184

(<.0001)

)0.100

(<.0001)

AdjROA

0.024

(<.0001)

)0.115

(<.0001)

)0.001

(0.878)

0.019

(<.0001)

)0.117

(<.0001)

)0.005

(0.262)

0.018

(0.001)

)0.040

(<.0001)

0.011

(0.028)

0.287

(<.0001)

)0.309

(<.0001)

0.001

(0.807)

SIZ

E0.018

(0.003)

)0.137

(<.0001)

)0.013

(0.031)

0.020

(<.0001)

)0.139

(<.0001)

)0.020

(<.0001)

0.017

(0.001)

)0.062

(<.0001)

0.018

(0.000)

0.256

(<.0001)

)0.281

(<.0001)

0.024

(<.0001)

0.029

(<.0001)

MTB

)0.057

(<.0001)

0.167

(<.0001)

)0.004

(0.529)

)0.059

(<.0001)

0.090

(<.0001)

)0.015

(0.002)

)0.059

(<.0001)

0.076

(<.0001)

)0.019

(0.000)

)0.075

(<.0001)

0.113

(<.0001)

0.037

(<.0001)

)0.125

(<.0001)

)0.096

(<.0001)

RETURN

)0.023

(0.000)

0.092

(<.0001)

0.009

(0.151)

)0.030

(<.0001)

0.038

(<.0001)

)0.004

(0.400)

)0.031

(<.0001)

0.020

(0.000)

)0.007

(0.167)

0.127

(<.0001)

)0.117

(<.0001)

)0.032

(<.0001)

0.006

(0.320)

0.003

(0.611)

0.272

(<.0001)

ZSCORE

0.001

(0.815)

)0.009

(0.160)

0.000

(0.965)

)0.001

(0.855)

)0.004

(0.365)

0.000

(0.933)

)0.001

(0.844)

)0.003

(0.545)

0.000

(0.941)

0.037

(<.0001)

)0.039

(<.0001)

0.00

(0.904)

0.029

(<.0001)

0.020

(0.001)

)0.035

(<.0001)

0.007

(0.279)

Notes:

Sample

consistsoffirm

-years

from

1988to

2002.RM

residualsare

estimatedfrom

(1)–(4).Two-tailed

p-values

presentedbelow

thecorrelationvalue.

Thevariablesare

defined

asfollows:

(Thetable

iscontinued

onthenextpage.)



TABLE5(C

ontinued)

R&D

RM


equalto


theR&D

model

1isin


otherwise

SG&A

RM


equalto


theSG&A

model

2isin


otherwise

ProductionRM


equalto


productionmodel

4isin


otherwise

BENCH


equalto

oneif(a)net

incomedivided

bytotalassetsisbetween0and0.01,or(b)thechangein

net

incomedivided

bytotalassetsbetween

t)1andtisbetween0and0.01,zero

otherwise

BEAT


equalto

oneif(a)net

incomedivided



net

incomedivided

bytotal

assetsbetweent


0.01and(c)BENCH

notequalto

one,

zero

otherwise

MISS


thatissetequalto

oneif(a)net

incomedivided

bytotalassetsisless

than

)0.01,or(b)thechangein

net

incomedivided

bytotalassets

betweent)1andtisless

than

)0.01and(c)BENCH,BEAT

orJUSTMISSisnotequalto

one,

zero

otherwise

JUSTMISS


equalto

oneif(a)net

incomedivided


)0.01butless

than0,or(b)thechangein

net

income

divided

bytotalassetsbetweent)1andtisgreaterthanorequalto

)0.01butless

than0and(c)BENCH

orBEAT

isnotequalto

one,

zero

otherwise

ROA

=incomebefore

extraordinary

item

sdivided

lagged

totalassets

AdjROA

=thedifference

betweenfirm

-specificROA

andthemedianROA


(two-digitSIC

)

SIZ

E=

thenaturallogarithm

oftotalassets

MTB

=themarket

valueofequitydivided

bythebookvalueofequity

RETURN

=size

adjusted

abnorm

alreturn

computedasthemonthly

buyandhold

raw

return

minusthemonthly

buyandhold

return

onasize

matched

decileportfolioof

firm

scompounded

over

12monthsoffiscalyeart

ZSCORE

=ameasure

offinancialhealthcomputedas:3.3*(N

etIncome t

⁄Assets t

)1)+

1.0*(Sales t

⁄Assets t

)1)+

1.4*(R

etained

Earnings t

⁄Assets t

)1)+

1.2*(W

orking

Capital t

⁄Assets t

)1)



Therefore, the coefficient of interest c5 represents the performance ofBENCH firms that use RM compared to non-RM MISS firms. Focusingon RM conditional on an earnings management incentive helps mitigate theeffects of alternative explanations and potential correlated omitted vari-ables. The uninteracted RM coefficient may proxy for managers’ attemptsto influence the output of the accounting system (i.e., RM) or some othermotivation omitted from the RM model. For example, a reduction in R&Drelative to other firms in the same year and industry (controlling for otherfactors) may reflect a manager attempting to influence the output of theaccounting system. However, it may also be picking up an omitted variable,such as a manager cutting the R&D budget when faced with decreasingreturns to R&D. In this case, decreasing returns to R&D may be negativelyassociated with future performance and the negative RM coefficient mayreflect the underlying economics of the firm and not the relation with realactivities manipulation.

Table 5 presents correlations for the variables in the future performanceregressions and a few variables appear to be highly correlated. In particular,the correlation between BENCH and BENCH*RM is around 0.40 for allthree RM samples. AdjROA is highly correlated with SIZE, MTB, andZSCORE, indicating the need to control for these variables in model 6.RETURN is highly correlated with MTB (0.27). The variance inflation fac-tors for the independent variables used in (6), for all three RM measures,are all less than 2.2 suggesting multicollinearity is likely not to be anissue.20

Table 6 presents the coefficient estimates for (6). I discuss theuntabulated results for t + 2 and t + 3 concurrent with discussing thet + 1 results reported in Table 6. With the exception of MTB, the controlvariables manifest predicted signs. The coefficient estimate on AdjROA issignificant and positive, indicating that current-period industry-adjustedROA is positively associated with future industry-adjusted ROA. RETURNis positive and significant consistent with Kothari and Sloan 1992. The firstcolumn of Table 6, panel A reports the results for the R&D RM sampleusing AdjROAt + 1 as the performance measure. The coefficient on BEATis 0.110, indicating that firms that beat the earnings benchmark by 0.01 ormore have incrementally higher AdjROAt + 1, ceteris paribus, than non-RM firms that miss the earnings benchmark by more than 0.01. On aver-age, AdjROAt + 1 for BEAT firms is )0.050 (c0 + c1) which is lower thanthe average reported in Table 4 (0.055) and this difference is mainly due tocontrolling for SIZE and lagged AdjROA. The coefficient on BENCH is0.056 (p-value < 0.001) and the coefficient on BENCH*RM is 0.031

20. Variance inflation factors (VIFs) are calculated using the R2 from the regression of that

particular independent variable on all the other independent variables. Higher VIFs are

indicative of collinearity problems. Greene (2000, 255–56) states, ‘‘as a rule of thumb,

for standardized data a VIF > 10 indicates harmful collinearity’’.



TABLE

6

Cross-sectionalregressionrelatingfuture

perform

ance

int+

1to

RM

Adj

RO

Atþ

1ðP

anel

A)

or

Adj

CFO

tþ1ðP

anel

BÞ¼

c 0þ

c 1B

EA

Ttþ

c 2JU

STM

ISS

tþ

c 3B

EN

CH

tþ

c 4R

Mtþ

c 5B

EN

CH� R

Mtþ

c 6R

OA

t

þc 7

SIZ

Etþ

c 8M

TB

tþ

c 9R

ETU

RN

tþ

c 10Z

SC

OR

Et�

1þ

e tþ

1

Panel

A:Industry-adjusted

return

onassets

Pred.

sign

R&D

sample

SG&A

sample

Production

sample

Aggregate

RM

sample

(R&D,SG&A,andProduction)

Intercept

))0.160()18.00)***

)0.117()25.45)***

)0.109()16.05)***

)0.132()15.13)***

BEATt

+0.110(20.72)***

0.091(34.89)***

0.075(14.96)***

0.089(14.74)***

JUSTMISSt

?0.051(6.71)***

0.039(5.93)***

0.032(5.33)***

0.045(5.30)***

BENCH

t?

0.056(8.28)***

0.042(7.87)***

0.035(6.82)***

0.042(5

99)***

RM

t?

0.008(1.60)

)0.023()9.24)***

)0.022()5.96)***

)0.037()7.51)***

BENCH

t*RM

t?

0.031(2.12)**

0.043(3.57)***

0.031(3.50)***

0.047(3.25)***

AdjROAt

+0.265(26.10)***

0.278(83.61)***

0.314(15.43)***

0.300(14.09)***

SIZ

Et

0.016(13.58)***

0.011(20.85)***

0.011(11.89)***

0.014(11.52)***

MTBt

+)0.005()5.65)***

)0.003()14.03)***

)0.003()4.31)***

)0.004()4.80)***

RETURN

t+

0.014(5.42)***

0.010(9.86)***

0.010(4.37)***

0.014(5.19)***

ZSCOREt

+0.000()0.37)

0.000()0.12)

0.003(1.33)

0.003(1.14)

Industry

dummies

Yes

Yes

Yes

Yes

Yeardummies

Yes

Yes

Yes

Yes

N23,041

36,501

31,855

20,701

R2

0.36

0.35

0.35

0.37

(Thetable

iscontinued

onthenextpage.)



TABLE

6(C

ontinued)

Panel

B:Industry-adjusted

cash

flow

from

operations

Pred.sign

R&D

sample

SG&A

sample

ProductionRM

Aggregate

RM

(R&D,

SG&A,andProduction)

Intercept

))0.154()19.28)***

)0.107()17.21)***

)0.099()15.46)***

)0.124()15.60)***

BEATt

+0.078(17.20)***

0.060(17.29)***

0.047(10.83)***

0.061(11.88)***

JUSTMISSt

?0.040(5.54)***

0.026(5.06)***

0.015(2.82)***

0.027(3.50)***

BENCH

t?

0.049(9.21)***

0.033(8.13)***

0.022(4.90)***

0.032(5.42)***

RM

t?

0.014(3.35)***

)0.026()7.30)***

)0.025()6.83)***

)0.039()8.22)***

BENCH

t*RM

t?

0.029(2.37)**

0.025(3.00)***

0.027(3.39)***

0.046(4.27)***

AdjROAt

+0.216(23.51)***

0.224(26.49)***

0.256(14.30)***

0.246(12.92)***

SIZ

Et

0.020(18.04)***

0.015(17.09)***

0.015(16.21)***

0.018(15.46)***

MTBt

+)0.003()4.49)***

)0.002()3.21)***

)0.002()3.31)***

)0.003()3.79)***

RETURN

t+

0.010(5.20)***

0.008(4.48)***

0.007(3.75)***

0.011(5.04)***

ZSCOREt

+0.000()0.53)

0.000()0.19)

0.002(1.27)

0.002(1.09)

Industry

dummies

Yes

Yes

Yes

Yes

Yeardummies

Yes

Yes

Yes

Yes

N22,977

36,410

31,778

20,645

R2

0.37

0.31

0.32

0.36

Notes:

*⁄**

⁄***representstatisticalsignificance

at10percent⁄5

percent⁄1

percentlevels,tw

o-tailed.t-testsin

parentheses.Sample

consistsof

firm

-years

from

1988to

2002.Thet-testsare

computedusingRoger’srobust

standard

errors

correctingforfirm

clusters.

Thevariablesare

defined

asfollows:

ROA

=incomebefore

extraordinary

item

sdivided

lagged

totalassets

(Thetable

iscontinued

onthenextpage.)



TABLE

6(C

ontinued)

AdjROA

=thedifference

betweenfirm

-specificROA

andthemedianROA


(two-digitSIC

)

CFO

=cash

flow

from

operationsdivided

bylagged

totalassets

AdjCFO

=thedifference

betweenfirm

-specificCFO

andthemedianCFO


(two-digitSIC

)

BENCH


equalto

oneif(a)net

incomedivided


0butless

than


net

incomedivided



0butless

than0.01,zero

otherwise

BEAT


equalto

oneif(a)net

incomedivided


0.01,or(b)the

changein

net

incomedivided



0.01and(c)BENCH

not

equalto

one,

zero

otherwise

JUSTMISS


equalto

oneif(a)net

incomedivided


)0.01butless

than0,or(b)thechangein

net

incomedivided



)0.01but

less

than0and(c)BENCH

orBEAT

isnotequalto

one,

zero

otherwise

R&D

RM


equalto


theR&D

model

1isin


otherwise

SG&A

RM


equalto


theSG&A

model

2isin


otherwise

ProductionRM


equalto


productionmodel

4isin


otherwise

Aggregate

RM


equalto

oneifthesum

oftheresidualsfrom

theR&D

model

1,SG&A

model

2,production

model

3multiplied

by

)1isin


otherwise

SIZ

E=

thenaturallogarithm

oftotalassets

MTB

=themarket

valueofequitydivided

bythebookvalueofequity

RETURN

=size

adjusted

abnorm

alreturnscomputedasthemonthly

buyandhold

raw

return

minusthemonthly

buyandhold

return

onasize

matched

decileportfoliooffirm

scompounded

over

12monthsoffiscalyeart

ZSCORE

=ameasure

offinancialhealthcomputedas:3.3*(N

etIncome t

⁄Assets t

)1)+

1.0*(Sales t

⁄Assets t

)1)+

1.4*(R

etained

Earnings t

⁄Assets t

)1)+

1.2*(W

orkingCapital t

⁄Assets t

)1)



(p-value < 0.05), suggesting that firms that just meet earnings benchmarksperform better on average than MISS or JUSTMISS (coefficient 0.051,p-value < 0.001) firms, but worse than BEAT firms (coefficient 0.110,p-value < 0.001), consistent with Bartov et al. 2002.

The coefficient on the interaction term (BENCH*RM) of 0.031 suggeststhat managers who engage in RM to just meet earnings benchmarks havebetter subsequent performance than non-RM MISS firms, which does notsupport Hypothesis 2. The average performance (ROAt + 1) of firms that justmeet the benchmark without engaging in RM, ceteris paribus, is )10.40 per-cent (c0 + c3) whereas the average performance of firms that just meet thebenchmark by engaging in RM is )6.50 percent (c0 + c3 + c4 + c5). Thep-value from a F-test of [(c4 + c5) = 0] is 0.027, suggesting that firms thatjust meet the benchmark by engaging in R&D RM have significantly higherindustry-adjusted ROA in t + 1 than non-RM BENCH firms. This result isconsistent with the joint signal — engaging in RM and just meeting the earn-ings benchmark — signaling superior future performance. In addition, theaverage performance of JUSTMISS firms is )0.109 (c0 + c2) indicating thatBENCH firms that engage in RM exhibit better subsequent performance thanfirms who just miss the earnings benchmarks. The results are robust tousing AdjROAt+2 and AdjROAt+3 as the future performance measure.The results are similar using AdjCFOt+1 (Table 6, panel B); for example,the coefficients on RM and BENCH*RM are positive and significant and thep-value from a F-test of [(c4 +c5) = 0] is 0.007.

The results for the SG&A sample are reported in the second column ofTable 6. The coefficients on the intercept, BEAT, and JUSTMISS are simi-lar to those of the R&D sample. The coefficient on RM is )0.023(p-value < 0.001), suggesting that firms that do not just meet the earningsbenchmark (i.e., non-BENCH) but engage in RM perform worse thannon-RM MISS firms. However, the coefficient on BENCH*RM is0.043 (p-value < 0.001), suggesting that managers of BENCH firms whoengage in RM have better subsequent performance compared to non-RMMISS firms. The results with respect to BENCH*RM are robust to usingAdjROAt+2 and AdjROAt+3 as the future performance measure. The aver-age performance of firms that just meet the benchmark without engaging inRM is )7.50 percent, whereas the average performance of firms that justmeet the benchmark by engaging in RM is )5.50 percent. The p-value froma F-test of [(c4 + c5) = 0] is 0.083, indicating that BENCH firms whoengage in SG&A RM have significantly higher industry-adjusted ROA int + 1 than non-RM BENCH firms. The results are robust using AdjROAt+2

and AdjROAt+3 as the performance measures. Table 6, panel B reports theresults using industry-adjusted CFO as the performance measure. Thecoefficient on BENCH*RM is 0.025 (p-value < 0.001), suggesting that man-agers of BENCH firms who engage in RM have higher subsequent CFOcompared to non-RM MISS firms. The average performance of BENCHfirms that do not engage in RM is )7.40 percent, whereas the average



performance of BENCH firms that engage in RM is )7.50 percent. Thedifference is not significant. While BENCH firms that engage in RMhave higher AdjCFOt+1 than non-RM MISS firms (coefficient 0.025,p-value < 0.001), it is not different from non-RM BENCH firms.

The results for production RM are presented in the third column inTable 6. The coefficient on the interaction term BENCH*RM is 0.031(p-value < 0.001). Untabulated results reveal that, in years t + 2 andt + 3, the coefficients on the interaction (BENCH*production RM) is 0.022(p-value < 0.02) and 0.019 (p-value < 0.08), respectively. The results aresimilar for future AdjCFO; however, year t + 3 is insignificant. The averageperformance of firms that just meet the benchmark without engaging inproduction RM is )7.40 percent, whereas the average performance of firmsthat just meet the benchmark by engaging in RM is )6.50 percent. Thep-value from a F-test of [(c4 + c5) = 0] is 0.4749. Therefore, the resultssuggest BENCH firms who engage in RM are associated with better perfor-mance in the subsequent three years compared to non-RM MISS firms butnot compared to non-RM BENCH firms.21

If firms engage in RM, they might engage in one or more types ofRM simultaneously; therefore, I aggregate the three RM measures shownto be associated with just meeting zero and last year’s earnings. The lastcolumn in Table 6 reports the results from the estimation of (6) with theaggregate measure. The results are similar to the individual measures.Overall, it appears that managers engage in RM to just meet earningsbenchmarks by cutting discretionary expense and using sales manipula-tion ⁄overproduction. The evidence presented in this section suggests thatusing RM to influence the output of the accounting system (i.e., to justmeet an earnings benchmark) is not opportunistic, but consistent withattaining benefits that allow the firm to perform better in the future or sig-naling future performance.22

6. Conclusion

This paper contributes to the body of literature examining the resourceallocation impact of earnings management. I examine four types of RM: (1)

21. Because production RM reflects two types of RM, overproduction to decrease COGS

expense and ⁄ or cutting prices or extending more lenient credit terms to boost sales, I

reestimate (6), excluding all nonmanufacturing firms. The coefficient on the interaction

term BENCH*RM is significantly positive in the subsequent three years for the AdjROA

and AdjCFO sample. Therefore, the results are robust to the manufacturing sample.

22. Because survivorship bias may influence the future performance results, I analyze the

rate and reason firms drop out of the BENCH and non-BENCH (MISS, JUSTMISS,

BEAT) samples. For each firm that drops out, I examine the delisting codes in CRSP

(delisting codes above 400 are classified as liquidation; delisting codes in the 200s are

classified as merger; all other codes are classified as other) to determine if there is a sig-

nificant difference in the firms that drop out of each sample. The firms appear to drop

out of the two samples at consistent rates and for consistent reasons; thus, I believe sur-

vivorship has a minimal effect on the results.



cutting discretionary investment of R&D to decrease expense, (2) cuttingdiscretionary investment of SG&A to decrease expense, (3) selling fixedassets to report gains, and (4) cutting prices or extending more lenient creditterms to boost sales and ⁄or overproduce to decrease COGS expense. First,I examine whether measures of these RM are associated with firms justmeeting two earnings benchmarks (zero and last year’s earnings). Second, Iassess the extent to which RM to meet earnings benchmarks is associatedwith future performance. The results indicate that after controlling for size,performance, and market-to-book, RM is positively associated with firmsjust meeting earnings benchmarks. Next, I find using RM to just meet earn-ings benchmarks is positively associated with future performance comparedto firms that do not use RM and miss the earnings benchmark by morethan 0.01. In addition, I find that firms that just meet earnings benchmarksby engaging in R&D or SG&A RM have significantly higher subsequentindustry-adjusted ROA than firms that do not engage in RM and just meetearnings benchmarks. In this setting, the results suggest earnings manage-ment via RM is not opportunistic, but consistent with managers attainingbenefits that allow better future performance or signaling.

This paper makes the following contributions. First, it contributes tothe literature on earnings management. By undertaking a comprehensiveexamination of four types of RM, this paper extends extant research investi-gating the consequences of earnings management. Although there are sev-eral studies documenting whether RM occurs in various situations, theexisting literature provides little evidence of the effect of RM on firms’ sub-sequent operating performance (with the exception of Bens et al. 2002).Without this type of analysis, it is difficult to determine whether managersuse RM, documented in prior literature, opportunistically. Second, thispaper contributes to the literature on earnings quality. Persistence of earn-ings is an important part of the ‘‘quality of earnings’’. In studies of financialstatement analysis, researchers are interested in how current or past earn-ings or earnings components aid in forecasting future earnings or cashflows, both of which are central inputs in valuation models. Examining theimplication of RM on performance is important given the significance offuture performance to the firm and its stakeholders. This paper shows thatusing empirical measures to identify firms that engage in RM is incremen-tally informative about future earnings.

References

Anderson, M., R. D. Banker, and S. N. Janakiraman. 2003. Are selling, gen-

eral, and administrative costs ‘‘sticky?’’ Journal of Accounting Research 41 (1):

47–63.

Baber, W., P. M. Fairfield, and J. A. Haggard. 1991. The effect of concern about

reported income on discretionary spending decisions: The case of research and

development. Accounting Review 66 (4): 818–29.



Barton, J., and P. Simko. 2002. The balance sheet as an earnings management con-

straint. Accounting Review 77 (Supplement): 1–27.

Bartov, E. 1993. The timing of asset sales and earnings manipulation. Accounting

Review 68 (4): 840–55.

Bartov, E., D. Givoly, and C. Hayn. 2002. The rewards to meeting or beating

earnings expectations. Journal of Accounting and Economics 33 (2): 173–204.

Bens, D., V. Nagar, and M. F. H. Wong. 2002. Real investment implications of

employee stock option exercises. Journal of Accounting Research 40 (2):

359–406.

Berger, P. 1993. Explicit and implicit tax effects of the R&D tax credit. Journal of

Accounting Research 31 (2): 131–71.

Burgstahler, D., and I. Dichev. 1997. Earnings management to avoid earnings

decreases and losses. Journal of Accounting and Economics 24 (2): 99–126.

Bushee, B. 1998. The influence of institutional investors on myopic R&D investment

behavior. Accounting Review 73 (3): 305–33.

Cheng, S. 2004. R&D expenditures and CEO compensation. Accounting Review 79

(2): 305–27.

Dechow, P., S. Kothari, and R. Watts. 1998. The relation between earnings and

cash flows. Journal of Accounting and Economics 25 (2): 133–69.

Dechow, P. M., and D. J. Skinner. 2000. Earnings management: Reconciling the

views of accounting academics, practitioners, and regulators. Accounting

Horizons 14 (2): 235–50.

Dechow, P., and R. Sloan. 1991. Executive incentives and the horizon problem.

Journal of Accounting and Economics 14 (1): 51–89.

Degeorge, F., J. Patel, and R. Zeckhauser. 1999. Earnings management to exceed

thresholds. Journal of Business 72 (1): 1–33.

Ewert, R., and A. Wagenhofer. 2005. Economic effects of tightening accounting

standards to restrict earnings management. Accounting Review 80 (4): 1101–25.

Fama, E., and J. Macbeth. 1973. Risk, return and equilibrium: Empirical tests.

Journal of Political Economy 81 (3): 607–36.

Fischer, P., and R. Verrecchia. 2000. Reporting bias. Accounting Review 75 (2):

229–45.

Graham, J., R. Harvey, and S. Rajgopal. 2005. The economic implications of

corporate financial reporting. Journal of Accounting and Economics 40 (1–3): 3–73.

Hayn, C. 1995. The information content of losses. Journal of Accounting and

Economics 20 (2): 125–49.

Herrmann, T., T. Inoue, and W. B. Thomas. 2003. The sale of assets to manage

earnings in Japan. Journal of Accounting Research 41 (1): 89–108.

Jacob, J., and B. Jorgensen. 2007. Earnings management and accounting income

aggregation. Journal of Accounting and Economics 43 (2 ⁄ 3): 369–90.Jones, J. 1991. Earnings management during import relief investigations. Journal of

Accounting Research 29 (2): 193–228.

Kothari, S., and R. Sloan. 1992. Information in prices about future earnings:

Implications for earnings response coefficients. Journal of Accounting and

Economics 15 (2): 143–72.



Lev, B. 2003. Corporate earnings: Fact and fiction. Journal of Economic Perspectives

17 (2): 27–50.

Mackie-Mason, J. 1990. Do taxes affect corporate financing decisions? Journal of

Finance 45 (5): 1471–93.

Matsumoto, D. 2002. Management’s incentives to avoid negative earnings surprises.

Accounting Review 77 (3): 483–514.

Matsunaga, S., and C. Park. 2001. The effect of missing a quarterly earnings

benchmark on the CEO’s annual bonus. Accounting Review 76 (3): 313–32.

Perry, S., and T. Williams. 1994. Earnings management preceding management

buyout offers. Accounting Review 18 (2): 157–80.

Petersen, M. A. 2009. Estimating standard errors in finance panel data sets:

Comparing approaches. Review of Financial Studies 22 (1): 435–80.

Roychowdhury, S. 2006. Earnings management through real activities manipulation.

Journal of Accounting and Economics 42 (3): 335–70.

Schipper, K. 1989. Earnings management. Accounting Horizons 3 (1): 91–102.

Subramanyam, K. R. 1996. The pricing of discretionary accruals. Journal of

Accounting and Economics 22 (1–3): 249–81.

Thomas, J. K., and H. Zhang. 2002. Inventory changes and future returns. Review

of Accounting Studies 7 (2–3): 163–87.

Trueman, B., and S. Titman. 1988. An explanation for accounting income

smoothing. Journal of Accounting Research 26 (Supplement): 127–39.

Watts, R., and J. Zimmerman. 1978. Towards a positive theory of the determination

of accounting standards. Accounting Review 53 (1): 112–34.

Zang, A 2007. Evidence on the tradeoff between real manipulation and accruals manip-

ulation. Working paper, University of Rochester.



Date post:	30-Apr-2020
Category:	Documents
Upload:	others
View:	4 times
Download:	0 times

The Relation Between Earnings Management Using Real ...€¦ · The Relation Between Earnings...

Documents